In the early 1990s, the number of computers, globally interconnected over the NSFNET, exceeded the number of computers on the ARPANET by far. Now ARPANET only represented one of hundreds of ARPA-internet-networks, and additionally it was a dinosaur, whose rate of adaptation was far beyond the rest of the internet.
Bob Kahn, the only remaining network-guru from ARPA had left the office in 1985 already. The current ARPA director didn’t show much interest towards network development. ARPANET itself, whose annual operational cost for ARPA was about 14 million dollar, appeared almost arthritic next to the much faster NSFNET. The ARPA management declared that ARPANET became a victim of its own success, and that it was time to shut it down eventually.
Mark Pullen, program manager for ARPA, was supposed to decommission the ARPANET. One after another, Pullen switched off IMPs and TIPs, which still made up the core of the original network. According to Pullen, a touch of mourning accompanied ARPANET’s retirement. Except for two computers, which went to MILNET, all remaining computers were handed to the local networks.
After growing for 20 years – from four hosts in the very beginning to finally 300.000 hosts – the end of the ARPANET had come.